From mboxrd@z Thu Jan 1 00:00:00 1970 From: Alasdair G Kergon Subject: Re: [dm-devel] [PATCH][RFC] dm: Do not open log and cow device read-write for read-only mappings Date: Tue, 15 Feb 2011 15:46:25 +0000 Message-ID: <20110215154625.GG5825@agk-dp.fab.redhat.com> References: <4D5A6EF4.3030905@redhat.com> <20110215124629.GF5825@agk-dp.fab.redhat.com> <20110215152033.GK3160@htj.dyndns.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Content-Disposition: inline In-Reply-To: <20110215152033.GK3160@htj.dyndns.org> Sender: linux-kernel-owner@vger.kernel.org To: Tejun Heo Cc: device-mapper development , Milan Broz , Jens Axboe , Tao Ma , linux-kernel@vger.kernel.org List-Id: dm-devel.ids On Tue, Feb 15, 2011 at 04:20:33PM +0100, Tejun Heo wrote: > Hmmm... but -EACCES is the correct one here. The device node itself > is rejecting RW access. There's no FS which is enforcing RO. Exactly:) If the filesystem permissions were what was blocking this (say r--) then I'd agree with EACCES. Interpret those man pages in the context of 'pathname refers to a block device not a file'. If it's EACCES, I just need to gain more privilege/capabilities and then repeat the system call and it could succeed. But EROFS tells me however much extra privilege I get it's going to make no difference. That's why I'm arguing EACCES is not a good error to return and EROFS is more appropriate. Alasdair